]> Cisco Systems, Inc.

Haarlerbergweg 13-19 Amsterdam 1101 CH Noord-Holland Netherlands +31 20 357 1752 klaas@cisco.com

Cisco Systems GmbH

Richtistrasse 7 Wallisellen CH-8304 ZH Switzerland +41 44 878 9200 lear@cisco.com

SJD AB

Hagagatan 24 Stockholm 113 47 SE simon@josefsson.org http://josefsson.org/

Security Assertion Markup Language (SAML) has found its usage on the Internet for Web Single Sign-On. Simple Authentication and Security Layer (SASL) and the Generic Security Service Application Program Interface (GSS-API) are application frameworks to generalize authentication. This memo specifies a SASL mechanism and a GSS-API mechanism for SAML 2.0 that allows the integration of existing SAML Identity Providers with applications using SASL and GSS-API.

Security Assertion Markup Language (SAML) 2.0 is a set of specifications that provide various means for a user to be identified to a relying party (RP) through the exchange of (typically signed) assertions issued by an identity provider (IdP). It includes a number of protocols, protocol bindings, and interoperability profiles designed for different use cases. Simple Authentication and Security Layer (SASL) is a generalized mechanism for identifying and authenticating a user and for optionally negotiating a security layer for subsequent protocol interactions. SASL is used by application protocols like IMAP, POP and XMPP. The effect is to make modular authentication, so that newer authentication mechanisms can be added as needed. This memo specifies just such a mechanism. The Generic Security Service Application Program Interface (GSS-API) provides a framework for applications to support multiple authentication mechanisms through a unified programming interface. This document defines a pure SASL mechanism for SAML, but it conforms to the new bridge between SASL and the GSS-API called GS2. This means that this document defines both a SASL mechanism and a GSS-API mechanism. The GSS-API interface is OPTIONAL for SASL implementers, and the GSS-API considerations can be avoided in environments that use SASL directly without GSS-API. As currently envisioned, this mechanism enables interworking between SASL and SAML in order to assert the identity of the user and other attributes to relying parties. As such, while servers (as relying parties) will advertise SASL mechanisms (including SAML), clients will select the SAML SASL mechanism as their SASL mechanism of choice. The SAML mechanism described in this memo aims to re-use the Web Browser SSO profile defined in section 4.1 of the SAML profiles 2.0 specification to the maximum extent and therefore does not establish a separate authentication, integrity and confidentiality mechanism. The mechanism assumes a security layer, such as Transport Layer Security (TLS), will continue to be used. This specification is appropriate for use when a browser instance is available. In the absence of a browser instance, SAML profiles that don't require a browser such as the Enhanced Client or Proxy profile (as defined in section 4.2 of the SAML profiles 2.0 specification may be used, but that is outside the scope of this specification. describes the interworking between SAML and SASL: this document requires enhancements to the Relying Party (the SASL server) and to the Client, as the two SASL communication end points, but no changes to the SAML Identity Provider are necessary. To accomplish this goal some indirect messaging is tunneled within SASL, and some use of external methods is made.

| Relying | / | Party | // | | // +-----------+ SAML/ // ^ HTTPS // +--|--+ // | S| | / S | A| | // A | M| | // S | L| | // L | | | // | | | | Client | | Provider | | | +------------+ +----------+ ]]>

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119. The reader is assumed to be familiar with the terms used in the SAML 2.0 specification.

Because this mechanism transports information that should not be controlled by an attacker, the SAML mechanism MUST only be used over channels protected by TLS, or over similar integrity protected and authenticated channels. In addition, when TLS is used the client MUST successfully validate the server certificate (, ) Note: An Intranet does not constitute such an integrity protected and authenticated channel!

While SAML itself is merely a markup language, its common use case these days is with HTTP or HTTPS and HTML. What follows is a typical flow: The browser requests a resource of a Relying Party (RP) (via an HTTP request). The Relying Party redirects the browser via an HTTP redirect (as described in Section 10.3 of ) to the Identity Provider (IdP) or an IdP discovery service. When it does so, it includes the following parameters: (1) an authentication request that contains the name of resource being requested, (2) a browser cookie, and (3) a return URL as specified in Section 3.1 of the SAML profiles 2.0 specification. The user authenticates to the IdP and perhaps authorizes the release of user attributes to the Relying Party. In its authentication response, the IdP redirects (via an HTTP redirect) the browser back to the RP with an authentication assertion (stating that the IdP vouches that the subject has successfully authenticated), optionally along with some additional attributes. The Relying Party now has sufficient identity information to approve access to the resource or not, and acts accordingly. The authentication is concluded. When considering this flow in the context of SASL, we note that while the Relying Party and the client both must change their code to implement this SASL mechanism, the IdP can remain untouched. The Relying Party already has some sort of session (probably a TCP connection) established with the client. However, it may be necessary to redirect a SASL client to another application or handler. The steps are as follows: The SASL server (Relying Party) advertises support for the SASL SAML20 mechanism to the client The client initiates a SASL authentication with SAML20 and sends a domain name that allows the SASL server to determine the appropriate IdP The SASL server transmits an authentication request encoded using a Uniform Resource Identifier (URI) as described in RFC 3986 and an HTTP redirect to the IdP corresponding to the domain The SASL client now sends an empty response, as authentication continues via the normal SAML flow and the SASL server will receive the answer to the challenge out-of-band from the SASL conversation. At this point the SASL client MUST construct a URL containing the content received in the previous message from the SASL server. This URL is transmitted to the IdP either by the SASL client application or an appropriate handler, such as a browser. Next the user authenticates to the IdP. The manner in which the end user is authenticated to the IdP and any policies surrounding such authentication is out of scope for SAML and hence for this draft. This step happens out of band from SASL. The IdP will convey information about the success or failure of the authentication back to the the SASL server (Relying Party) in the form of an Authentication Statement or failure, using a indirect response via the client browser or the handler (and with an external browser client control should be passed back to the SASL client). This step happens out of band from SASL. The SASL Server sends an appropriate SASL response to the client, along with an optional list of attributes Please note: What is described here is the case in which the client has not previously authenticated. It is possible that the client already holds a valid SAML authentication token so that the user does not need to be involved in the process anymore, but that would still be external to SASL. This is classic Web Single Sign-On, in which the Web Browser client presents the authentication token (cookie) to the RP without renewed user authentication at the IdP. With all of this in mind, the flow appears as follows in :

-----(1)----->| | Advertisement | | | |<-----(2)-----<| | Initiation | | | |>-----(3)----->| | Authentication Request | | | |<-----(4)-----<| | Empty Response | | | | |< - -(5,6) - ->| Client<>IDP | | | Authentication | | | |<- - - - - - - - - - -(7)- - -| Authentication Statement | | | |>-----(8)----->| | SASL completion with | | | status | | | ----- = SASL - - - = HTTP or HTTPS (external to SASL) ]]>

This section specifies the details of the SAML SASL mechanism. See section 5 of for what is described here. The name of this mechanism is "SAML20". The mechanism is capable of transferring an authorization identity (via the "gs2-header"). The mechanism does not offer a security layer. The mechanism is client-first. The first mechanism message from the client to the server is the "initial-response". As described in , if the application protocol does not support sending a client-response together with the authentication request, the server will send an empty server-challenge to let the client begin. The second mechanism message is from the server to the client, containing the SAML "authentication-request". The third mechanism message is from client to the server, and is the fixed message consisting of "=" (i.e., an empty response). The fourth mechanism message is from the server to the client, indicating the SASL mechanism outcome.

A client initiates a "SAML20" authentication with SASL by sending the GS2 header followed by the authentication identifier (message 2 in ) and is defined as follows:

The "gs2-header" is used as follows: - The "gs2-nonstd-flag" MUST NOT be present. - The "gs2-cb-flag" MUST be set to "n" because channel binding data cannot be integrity protected by the SAML negotiation. (Note: In theory channel binding data could be inserted in the SAML flow by the client and verified by the server, but that is currently not supported in SAML.) - The "gs2-authzid" carries the optional authorization identity as specified in (not to be confused with the IdP-Identifier). Domain name is specified in . A domain name is either a "traditional domain name" as described in or an "internationalized domain name" as described in . Clients and servers MUST treat the IdP-Identifier as a domain name slot . They also SHOULD support internationalized domain names (IDNs) in the Idp-Identifier field; if they do so, all of the domain name's labels MUST be A-labels or NR-LDH labels , if necessary internationalized labels MUST be converted from U-labels to A-labels by using the Punycode encoding for A-labels prior to sending them to the SASL-server as described in the protocol specification for Internationalized Domain Names in Applications.

The SASL Server transmits to the SASL client a URI that redirects the SAML client to the IdP (corresponding to the domain that the user provided), with a SAML authentication request as one of the parameters (message 3 in ) in the following way:

URI is specified in and is encoded according to Section 3.4 (HTTP Redirect) of the SAML bindings 2.0 specification. The SAML authentication request is encoded according to Section 3.4 (Authentication Request) of the SAML core 2.0 specification. Should the client support Internationalized Resource Identifiers (IRIs) it MUST first convert the IRI to a URI before transmitting it to the server . Note: The SASL server may have a static mapping of domain to corresponding IdP or alternatively a DNS-lookup mechanism could be envisioned, but that is out-of-scope for this document. Note: While the SASL client MAY sanity check the URI it received, ultimately it is the SAML IdP that will be validated by the SAML client which is out-of-scope for this document. The client then sends the authentication request via an HTTP GET (sent over a server-authenticated TLS channel) to the IdP, as if redirected to do so from an HTTP server and in accordance with the Web Browser SSO profile, as described in section 3.1 of SAML profiles 2.0 specification (message 5 and 6 in ). The client handles both user authentication to the IdP and confirmation or rejection of the authentiation of the RP (out-of-scope for this document). After all authentication has been completed by the IdP, the IdP will send a redirect message to the client in the form of a URI corresponding to the Relying Party as specified in the authentication request ("AssertionConsumerServiceURL") and with the SAML response as one of the parameters (message 7 in ). Please note: this means that the SASL server needs to implement a SAML Relying Party. Also, the SASL server needs to correlate the session it has with the SASL client with the appropriate SAML authentication result. It can do so by comparing the ID of the SAML authentication request it has issued with the one it receives in the SAML authentication statement.

The SASL server (in its capacity as a SAML Relying Party) now validates the SAML authentication response it received from the SAML client via HTTP or HTTPS. The outcome of that validation by the SASL server constitutes a SASL mechanism outcome, and therefore (as stated in ) SHALL be used to set state in the server accordingly, and it SHALL be used by the server to report that state to the SASL client as described in Section 3.6 (message 8 in ).

This section and its sub-sections are not required for SASL implementors, but this section MUST be observed to implement the GSS- API mechanism discussed below. This section specify a GSS-API mechanism that when used via the GS2 bridge to SASL behaves like the SASL mechanism defined in this document. Thus, it can loosely be said that the SAML SASL mechanism is also a GSS-API mechanism. The SAML user takes the role of the GSS-API Initiator and the SAML Relying Party takes the role of the GSS-API Acceptor. The SAML Identity Provider does not have a role in GSS-API, and is considered an internal matter for the SAML mechanism. The messages are the same, but a) the GS2 header on the client's first message and channel binding data is excluded when SAML is used as a GSS-API mechanism, and b) the RFC2743 section 3.1 initial context token header is prefixed to the client's first authentication message (context token). The GSS-API mechanism OID for SAML is OID-TBD (IANA to assign: see IANA considerations). SAML20 security contexts MUST have the mutual_state flag (GSS_C_MUTUAL_FLAG) set to TRUE. SAML does not support credential delegation, therefore SAML security contexts MUST have the deleg_state flag (GSS_C_DELEG_FLAG) set to FALSE. The mutual authentication property of this mechanism relies on successfully comparing the TLS server identity with the negotiated target name. Since the TLS channel is managed by the application outside of the GSS-API mechanism, the mechanism itself is unable to confirm the name while the application is able to perform this comparison for the mechanism. For this reason, applications MUST match the TLS server identity with the target name, as discussed in . More precisely, to pass identity validation the client uses the securely negotiated targ_name as the reference identifier and match it to the DNS-ID of the server certificate, and MUST reject the connection if there is a mismatch. For compatibility with deployed certificate hierarchies, the client MAY also perform a comparison with the CN-ID when there is no DNS-ID present. Wildcard matching is permitted. The targ_name reference identifier is a "traditional domain names" thus the comparison is made using case-insensitive ASCII comparison. The SAML mechanism does not support per-message tokens or GSS_Pseudo_random.

SAML supports standard generic name syntaxes for acceptors such as GSS_C_NT_HOSTBASED_SERVICE (see , Section 4.1). SAML supports only a single name type for initiators: GSS_C_NT_USER_NAME. GSS_C_NT_USER_NAME is the default name type for SAML. The query, display, and exported name syntaxes for SAML principal names are all the same. There are no SAML-specific name syntaxes -- applications should use generic GSS-API name types such as GSS_C_NT_USER_NAME and GSS_C_NT_HOSTBASED_SERVICE (see , Section 4). The exported name token does, of course, conforms to , Section 3.2.

Suppose the user has an identity at the SAML IdP saml.example.org and a Jabber Identifier (JID) "somenode@example.com", and wishes to authenticate his XMPP connection to xmpp.example.com. The authentication on the wire would then look something like the following: Step 1: Client initiates stream to server:

]]>

Step 2: Server responds with a stream tag sent to client:

]]>

Step 3: Server informs client of available authentication mechanisms:

DIGEST-MD5 PLAIN SAML20 ]]>

Step 4: Client selects an authentication mechanism and provides the initial client response containing the according to the definition in Section 4 ofBASE64 encoded gs2-header and domain:

biwsZXhhbXBsZS5vcmc ]]>

The decoded string is: n,,example.org Step 5: Server sends a BASE64 encoded challenge to client in the form of an HTTP Redirect to the SAML IdP corresponding to example.org (https://saml.example.org) with the SAML Authentication Request as specified in the redirection url:

The decoded challenge is:

Where the decoded SAMLRequest looks like:

https://xmpp.example.com urn:oasis:names:tc:SAML:2.0:ac:classes:PasswordProtectedTransport ]]>

Note: the server can use the request ID (_bec424fa5103428909a30ff1e31168327f79474984) to correlate the SASL session with the SAML authentication. Step 5 (alternative): Server returns error to client if no SAML Authentication Request can be constructed:

]]>

Step 6: Client sends the empty response to the challenge encoded as a single =:

= ]]>

The following steps between brackets are out of scope for this document but included to better illustrate the entire flow. [The client now sends the URL to a browser instance for processing. The browser engages in a normal SAML authentication flow (external to SASL), like redirection to the Identity Provider (https://saml.example.org), the user logs into https://saml.example.org, and agrees to authenticate to xmpp.example.com. A redirect is passed back to the client browser who sends the AuthN response to the server, containing the subject-identifier as an attribute. If the AuthN response doesn't contain the JID, the server maps the subject-identifier received from the IdP to a JID] Step 7: Server informs client of successful authentication:

]]>

Step 7 (alt): Server informs client of failed authentication:

]]>

Please note: line breaks were added to the base64 for clarity.

The following describes an IMAP exchange. Lines beginning with 'S:' indicate data sent by the server, and lines starting with 'C:' indicate data sent by the client. Long lines are wrapped for readability.

S: * OK IMAP4rev1 C: . CAPABILITY S: * CAPABILITY IMAP4rev1 STARTTLS S: . OK CAPABILITY Completed C: . STARTTLS S: . OK Begin TLS negotiation now C: . CAPABILITY S: * CAPABILITY IMAP4rev1 AUTH=SAML20 S: . OK CAPABILITY Completed C: . AUTHENTICATE SAML20 S: + C: biwsZXhhbXBsZS5vcmc S: + aHR0cHM6Ly9zYW1sLmV4YW1wbGUub3JnL1NBTUwvQnJvd3Nlcj9TQU1M UmVxdWVzdD1QSE5oYld4d09rRg0KMWRHaHVVbVZ4ZFdWemRDQjRiV3h1Y3pwe llXMXNjRDBpZFhKdU9tOWhjMmx6T201aGJXVnpPblJqT2xOQg0KVFV3Nk1pNH dPbkJ5YjNSdlkyOXNJZzBLSUNBZ0lFbEVQU0pmWW1Wak5ESTBabUUxTVRBek5 ESTRPVEE1WQ0KVE13Wm1ZeFpUTXhNVFk0TXpJM1pqYzVORGMwT1RnMElpQlda WEp6YVc5dVBTSXlMakFpRFFvZ0lDQWdTWA0KTnpkV1ZKYm5OMFlXNTBQU0l5T URBM0xURXlMVEV3VkRFeE9qTTVPak0wV2lJZ1JtOXlZMlZCZFhSb2JqMA0KaV ptRnNjMlVpRFFvZ0lDQWdTWE5RWVhOemFYWmxQU0ptWVd4elpTSU5DaUFnSUN CUWNtOTBiMk52YkVKcA0KYm1ScGJtYzlJblZ5YmpwdllYTnBjenB1WVcxbGN6 cDBZenBUUVUxTU9qSXVNRHBpYVc1a2FXNW5jenBJVg0KRlJRTFZCUFUxUWlEU W9nSUNBZ1FYTnpaWEowYVc5dVEyOXVjM1Z0WlhKVFpYSjJhV05sVlZKTVBRME tJQw0KQWdJQ0FnSUNBaWFIUjBjSE02THk5dFlXbHNMbVY0WVcxd2JHVXVZMjl 0TDFOQlRVd3ZRWE56WlhKMGFXOQ0KdVEyOXVjM1Z0WlhKVFpYSjJhV05sSWo0 TkNpQThjMkZ0YkRwSmMzTjFaWElnZUcxc2JuTTZjMkZ0YkQwaQ0KZFhKdU9tO WhjMmx6T201aGJXVnpPblJqT2xOQlRVdzZNaTR3T21GemMyVnlkR2x2YmlJK0 RRb2dJQ0FnSQ0KR2gwZEhCek9pOHZlRzF3Y0M1bGVHRnRjR3hsTG1OdmJRMEt JRHd2YzJGdGJEcEpjM04xWlhJK0RRb2dQSA0KTmhiV3h3T2s1aGJXVkpSRkJ2 YkdsamVTQjRiV3h1Y3pwellXMXNjRDBpZFhKdU9tOWhjMmx6T201aGJXVg0Ke k9uUmpPbE5CVFV3Nk1pNHdPbkJ5YjNSdlkyOXNJZzBLSUNBZ0lDQkdiM0p0WV hROUluVnlianB2WVhOcA0KY3pwdVlXMWxjenAwWXpwVFFVMU1Pakl1TURwdVl XMWxhV1F0Wm05eWJXRjBPbkJsY25OcGMzUmxiblFpRA0KUW9nSUNBZ0lGTlFU bUZ0WlZGMVlXeHBabWxsY2owaWVHMXdjQzVsZUdGdGNHeGxMbU52YlNJZ1FXe HNiMw0KZERjbVZoZEdVOUluUnlkV1VpSUM4K0RRb2dQSE5oYld4d09sSmxjWF ZsYzNSbFpFRjFkR2h1UTI5dWRHVg0KNGRBMEtJQ0FnSUNCNGJXeHVjenB6WVc xc2NEMGlkWEp1T205aGMybHpPbTVoYldWek9uUmpPbE5CVFV3Ng0KTWk0d09u QnliM1J2WTI5c0lpQU5DaUFnSUNBZ0lDQWdRMjl0Y0dGeWFYTnZiajBpWlhoa FkzUWlQZzBLSQ0KQ0E4YzJGdGJEcEJkWFJvYmtOdmJuUmxlSFJEYkdGemMxSm xaZzBLSUNBZ0lDQWdlRzFzYm5NNmMyRnRiRA0KMGlkWEp1T205aGMybHpPbTV oYldWek9uUmpPbE5CVFV3Nk1pNHdPbUZ6YzJWeWRHbHZiaUkrRFFvZ0lDQQ0K Z0lDQjFjbTQ2YjJGemFYTTZibUZ0WlhNNmRHTTZVMEZOVERveUxqQTZZV002W TJ4aGMzTmxjenBRWVhOeg0KZDI5eVpGQnliM1JsWTNSbFpGUnlZVzV6Y0c5eW RBMEtJQ0E4TDNOaGJXdzZRWFYwYUc1RGIyNTBaWGgwUQ0KMnhoYzNOU1pXWSt EUW9nUEM5ellXMXNjRHBTWlhGMVpYTjBaV1JCZFhSb2JrTnZiblJsZUhRK0lB MEtQQw0KOXpZVzFzY0RwQmRYUm9ibEpsY1hWbGMzUSs= C: S: . OK Success (tls protection)

The decoded challenge is:

Where the decoded SAMLRequest looks like:

https://xmpp.example.com urn:oasis:names:tc:SAML:2.0:ac:classes:PasswordProtectedTransport ]]>

This section addresses only security considerations associated with the use of SAML with SASL applications. For considerations relating to SAML in general, the reader is referred to the SAML specification and to other literature. Similarly, for general SASL Security Considerations, the reader is referred to that specification.

This mechanism is vulnerable to man-in-the-middle and tunneling attacks unless a client always verifies the server identity before proceeding with authentication (see ). Typically TLS is used to provide a secure channel with server authentication.

As specified in , the server is responsible for binding credentials to a specific authorization identity. It is therefore necessary that only specific trusted IdPs be allowed. This is typical part of SAML trust establishment between Relying Parties and IdP.

The IdP is aware of each Relying Party that a user logs into. There is nothing in the protocol to hide this information from the IdP. It is not a requirement to track the visits, but there is nothing that prohibits the collection of information. SASL server implementers should be aware that SAML IdPs will be able to track - to some extent - user access to their services.

It is possible for Relying Parties to link data that they have collected on the users. By using the same identifier to log into every Relying Party, collusion between Relying Parties is possible. In SAML, targeted identity was introduced. Targeted identity allows the IdP to transform the identifier the user typed in to a Relying Party specific opaque identifier. This way the Relying Party would never see the actual user identifier, but a randomly generated identifier.

Security issues inherent in GSS-API (RFC 2743) and GS2 (RFC 5801) apply to the SAML GSS-API mechanism defined in this document. Further, and as discussed in section 4, proper TLS server identity verification is critical to the security of the mechanism.

The IANA is requested to register the following SASL profile: SASL mechanism profile: SAML20 Security Considerations: See this document Published Specification: See this document For further information: Contact the authors of this document. Owner/Change controller: the IETF Intended usage: COMMON Note: None

The IANA is further requested to assign a new entry for this GSS mechanism in the sub-registry for SMI Security for Mechanism Codes, whose prefix is iso.org.dod.internet.security.mechanisms (1.3.6.1.5.5) and to reference this specification in the registry.

&RFC1035; &RFC2818; &RFC3492; &RFC3986; &RFC3987; &RFC2119; &RFC2616; &RFC2743; &RFC4422; &RFC5056; &RFC5246; &RFC5280; &RFC5890; &RFC5891; &RFC6125; &RFC5801; &SAML20; &SAML20BIND; &SAML20PROF; &W3C.REC-html401-19991224; &RFC1939; &RFC3501; &RFC4648; &RFC6120;

The authors would like to thank Scott Cantor, Joe Hildebrand, Josh Howlett, Leif Johansson, Thomas Lenggenhager, Diego Lopez, Hank Mauldin, RL 'Bob' Morgan, Stefan Plug and Hannes Tschofenig for their review and contributions.

This section to be removed prior to publication. 09 Fixed text per IESG review 08 Fixed text per Gen-Art review 07 Fixed text per comments Alexey Melnikov 06 Fixed text per AD comments 05 Fixed references per ID-nits 04 Added request for IANA assignment, few text clarifications 03 Number of cosmetic changes, fixes per comments Alexey Melnikov 02 Changed IdP URI to domain per Joe Hildebrand, fixed some typos 00 WG -00 draft. Updates GSS-API section, some fixes per Scott Cantor 01 Added authorization identity, added GSS-API specifics, added client supplied IdP 00 Initial Revision.